Computational Miniature Mesoscope V2: A deep learning-augmented miniaturized microscope for single-shot 3D high-resolution fluorescence imaging

Computational Miniature Mesoscope (CM2) is a recently developed computational imaging system that enables single-shot 3D imaging across a wide field-of-view (FOV) using a compact optical platform. In this work, we present CM2 V2 - an advanced CM2 system that integrates novel hardware improvements and a new deep learning reconstruction algorithm. The platform features a 3D-printed freeform LED collimator that achieves ~80$\%$ excitation efficiency - a ~3x improvement over our V1 design, and a hybrid emission filter design that improves the measurement contrast by >5x. The new computational pipeline includes an accurate and computationally efficient 3D linear shift-variant (LSV) forward model and a novel multi-module CM2Net deep learning model. As compared to the model-based deconvolution in our V1 system, CM2Net achieves ~8x better axial localization and ~1400x faster reconstruction speed. In addition, CM2Net consistently achieves high detection performance and superior axial localization across a wide FOV at a variety of conditions. Trained entirely on our 3D-LSV simulator generated training data set, CM2Net generalizes well to real experiments. We experimentally demonstrate that CM2Net achieves accurate 3D reconstruction of fluorescent emitters across a ~7-mm FOV and 800-$\mu$m depth, and provides ~7-$\mu$m lateral and ~25-$\mu$m axial resolution. We anticipate that this simple and low-cost computational miniature imaging system will be impactful to many large-scale 3D fluorescence imaging applications.